The present invention generally relates to photoplethysmographic measurement systems, and, in particular, to a method and system for improving photoplethysmographic analyte measurements by preprocessing measurement data to compensate for measurement artifact such as by de-weighting motion-contaminated data.
In the field of photoplethysmography, light pulses from different portions of the electromagnetic spectrum are used to noninvasively determine various blood analyte related values, such as blood oxygen saturation, in test subjects. Typically, photoplethysmographic measurement systems, such as pulse oximeters, include a probe for releasably attaching to the tip of a patient""s appendage (e.g., a finger, earlobe or the nasal septum). The probe directs light signals into the appendage where the probe is attached. Some portion of the light is absorbed by the tissue and a remaining portion of the light passes through patient tissue. The intensity of light passing through the tissue is monitored by a sensor typically contained in the probe. The intensity related signals produced by the sensor are used to compute blood analyte related values.
During a medical examination, measurements such as blood oxygen saturation levels computed by the pulse oximeter can be distorted by various factors including movement of the appendage where the probe is attached. For example, the movement of the patient may affect source/detector alignment, ambient light levels, blood volume fluctuation or other factors, thus resulting in potential measurement errors. Some instrument designs have attempted to address such artifact through the use of hardware such as filters to screen the detector signals. However, such approaches have generally had a limited ability to distinguish artifact components of the signal from desired information, resulting in loss of information and/or admission of substantial artifact into the data used for calculations.
There is a particular need for a photoplethysmographic measurement instrument that is capable of accurately analyzing the quality of output signals produced by an instrument sensor so that appropriate steps may be taken to improve photoplethysmographic analyte related measurements computed by the instrument. In particular, there is a need for a system with improved ability to distinguish artifact from desired information and to preprocess the data based on such distinction so as to enhance overall instrument performance.
The present invention is directed to a system and corresponding method for use in a pulse oximeter to improve the way in which data output by a pulse oximeter sensor, such as data representing a pulse waveform, is preprocessed prior to the computation of blood oxygen saturation levels. The present system quantifies artifactual components (e.g., motion artifact) contained in a set of data aggregated over a certain time interval (e.g., one or more pulse cycles or portions thereof or simply a given predetermined time interval) and preprocesses the set of data such that the effect of artifactual components is reduced. The blood oxygen saturation levels may be computed based on multiple sets of preprocessed data so as to improve the accuracy of the pulse oximetry measurements.
In accordance with one aspect of the present invention, an apparatus for evaluating the reliability of output signals produced by a pulse oximeter sensor is provided. Electrical output signals produced by the sensor may be captured at a defined interval. Each captured set of signals may be analyzed independently to estimate a degree of artifactual component reflected in the set of output signals. Unreliable sets of output signals can be addressed in a variety of ways. In one embodiment, a weight metric may be generated which accurately reflects the degree of artifactual component in the output signals. The weight metric may then be used to relatively emphasize or de-emphasize each set of output signals in proportion to the degree of artifactual component estimated such that the sets of output signals used in the blood analyte computation are dominated by output signals with a relatively small artifactual component. In another embodiment, sets of output signals that are deemed to be unreliable, i.e., the degree of artifactual component associated therewith exceeds a certain threshold, may be excluded from the blood analyte computation.
In accordance with a related aspect of the present invention, motion affected data is identified based on a numerical or mathematical analysis of the data. The electrical output signals generated by the photodetector may be separated into red and infrared data by a hardware or software based demultiplexer. The reliability of output signals may be evaluated on the basis of data points derived from separated red and infrared pulse waveforms. The motion contained in output signals produced by the pulse oximetry sensor may be quantified based on spread of the data points, e.g., from a regression model such as a best-fit linear regression line. In one embodiment, the red and infrared pulse waveforms are used to acquire differential absorption values dAx and dAy respectively. Each pulse cycle is represented by a set of data points, where each data point represents a pair of corresponding differential absorption values dAx and dAy obtained at a specific point in time within each pulse cycle. A principal component analysis (xe2x80x9cPCSxe2x80x9d) may be performed on each set of data points to estimate an amount of motion affecting the data. First and second principal components derived from the PCA account for different amounts of variation among the set of data points. The first principal component may describe the cluster of data points disposed along a longitudinal axis and the second principal component may describe the amount of spread of the data points with respect to the longitudinal axis. An amount of motion associated with a set of data points may be approximated by multiplying the RMS (Root Mean Square) variation along the axis of the first principal component by the RMS variation along the axis of the second principal component.
According to another related aspect of the invention, a method is provided for distinguishing motion artifact from the plethysmographic signal. It has been recognized that the artifact due to motion can be distinguished from other data anomalies because motion affects tend to be relatively evenly distributed over multiple channels (a channel being the electronic or digital data received from any given emitter in a photoplethysmographic system). Accordingly, motion effects can be identified and quantified by mathematically accounting for a corresponding bias or tendency in the data. For example, if two channels of data are plotted against one another over a set of samples, motion effects tend to be reflected by a slant, e.g., an approximately 45-degree bias in the data spread since some common types of motion tend to induce similar amounts of artifact into each channel (actually into the dA""s calculated from the channel data), while the plethysmographic signal will tend to spread the data at some other angle determined by blood analyte levels. It will be appreciated that such effects may be identified mathematically rather than graphically. In this regard, each data point generally includes a plethysmographic content and a motion content. The motion content of a data point may be represented by a shift of the data point at a 45-degree angle, the amount of shift indicating the degree of motion affecting the data point. Simultaneously, the data will also be spread along some other angle which is determined by blood analyte levels. The random spreading of points in two directions expands the area occupied by the points. Thus, motion artifact present in a data point may be quantified based on the area filled by the spreading of the points. Typically, a set of data points may be characterized by a parallelogram, the boundary of which encompasses a majority of data points. The amount of motion associated with such set of data points may be estimated by approximating the area of the parallelogram which outlines the data points.
According to another aspect of the invention, one or more buffers or data storage units may be utilized to temporarily store data generated by a pulse oximeter so that the data may be accessed and manipulated by processing modules to improve the accuracy of analyte related computation such as an oxygen saturation level computation. In a preferred embodiment, dual buffer system architecture is utilized. A first buffer is used to temporarily store a set of data representing at least one pulse cycle or other time period of data from a pulse oximetry sensor. A first processing module accesses the raw data stored in the first buffer to preprocess the raw data and yield preprocessed data. The preprocessed data can then be used, on a set-by-set or aggregated basis, to make the desired analyte related computations. For example, each set of preprocessed data may be transferred into a second buffer which is adapted to temporarily store multiple sets of preprocessed data most recently received from the first processing module.
The dual buffer system architecture of the present invention may be used for various purposes. For example, the data output by the sensor may be processed to eliminate, de-weight, or otherwise address data components that are not representative of the desired analyte related values, such as electronic noise, environmental noise, motion, or other artifact. In one embodiment, the first processing module may include a combination of a motion estimator and a weight application module. The motion estimator quantifies an amount of motion associated with the raw data stored in the first buffer by performing a principal component analysis on differential absorption values derived from the output signals. Based on the amount of motion estimated, the weight application module is configured to associate a weight with the raw data in the first buffer and transfer the weighted data to the second buffer. A second processing module may then assess multiple pulse cycles of improved data accumulated in the second buffer to perform its oxygen saturation level computation.
According to a further aspect of the invention, buffered or preprocessed data may be utilized to reduce the effect of artifactual component, such as motion artifact, in raw signal output by a pulse oximeter. A quantity of motion artifact identified in a set of input data, representing raw signals produced by a pulse oximeter sensor during a defined time period, may be utilized to selectively emphasize and/or de-emphasize the set of data based upon the quality of the data. In order to improve the accuracy of blood oxygen saturation computed by the pulse oximeter, the effect of data sets that are contaminated by an unacceptable amount of artifactual component may be substantially minimized or eliminated from the final computation thereof.
In one embodiment, a threshold value is used to distinguish between acceptable data (relatively free of motion) and unacceptable data (affected by motion). Data collected over each pulse cycle is considered in the blood analyte computation if the degree of motion associated therewith is below a threshold value. Otherwise, if the degree of motion exceeds the threshold, the data is discarded from further consideration in the final computation. In a preferred embodiment, weighting is used to either emphasize or de-emphasize data depending on the quantity of motion associated with the data. A quantity of motion content is estimated for each set of data stored in the first buffer. Then, a weight parameter is assigned to each set of data such that a motion-affected set of data may be relatively de-emphasized and/or a motion-free set of data is relatively emphasized. The blood analyte levels may be computed by performing a linear regression analysis on the weighted sets of data collected over multiple pulse cycles. In this regard, the slope of the best-fit line equation derived from the linear regression analysis provides an accurate blood oxygen level since the weighted sets of data are dominated by those with relative low motion content.
According to yet another aspect of the invention, processing logic may be utilized to compensate for motion-affected data. The process of estimating a quantity of motion and associating a weight parameter in proportion to the quantity of motion estimated may be embodied in the form of a software program or an executable set of instructions running on a processor. For example, a stream of pulse waveforms provided by the photodetector may be processed by a software routine configured to determine the quantity of motion artifact present in the raw data and manipulate the raw data in such a way as to produce an improved preprocessed data which more accurately reflects the actual blood oxygen saturation levels by reducing the effect of or eliminating raw data contaminated by undesirable artifactual content, i.e., due to movement of the patient.